IP vs IFCONFIG (was Re: Network card recommendation)
On Wed, 24 Aug 2005, Daniel L. Miller wrote: My firewall is continuing to run reliably - of course. I have noticed that on of the 3com NICs, I have a single RX overrun reported. Just one - no errors, and no increases in the overrun number. :00:0e.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 74) :00:0f.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 74) and - you didn't ask - Debian pre-compiled kernel 2.6.12-1-686 That said, once-every-year errors like these are not worth bothering with. Maybe something stalled your kernel and the buffer overloaded for that packet. It's curious - cuz it only happens during boot, and just the one overrun. New information on this. My firewall has been up now for some time - 13 days - and no further errors/overruns have been reported. Since I'm starting to learn the "ip" command, I just gave the "statistics" option a try. No overruns are reported on any of the network interfaces - yet the "ifconfig" command still shows (1) overrun on the one interface. Why!? -- Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Henrique de Moraes Holschuh wrote: On Wed, 24 Aug 2005, Daniel L. Miller wrote: My firewall is continuing to run reliably - of course. I have noticed that on of the 3com NICs, I have a single RX overrun reported. Just one - no errors, and no increases in the overrun number. Which 3com model exactly? (lspci output, please). :00:0e.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 74) :00:0f.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 74) and - you didn't ask - Debian pre-compiled kernel 2.6.12-1-686 That said, once-every-year errors like these are not worth bothering with. Maybe something stalled your kernel and the buffer overloaded for that packet. It's curious - cuz it only happens during boot, and just the one overrun. -- Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
On Wed, 24 Aug 2005, Daniel L. Miller wrote: > My firewall is continuing to run reliably - of course. I have noticed > that on of the 3com NICs, I have a single RX overrun reported. Just one > - no errors, and no increases in the overrun number. Which 3com model exactly? (lspci output, please). That said, once-every-year errors like these are not worth bothering with. Maybe something stalled your kernel and the buffer overloaded for that packet. I got a lot of overrun errors (plus frame errors) in a machine that has five 3c905B. Before kernel 2.6.11, I would get a ton of "too much work on interrupt" errors as well. This is the reason why I prefer the 3c905CX (slightly better behaviour) and Intel e1000 (I never managed to cause it to drop packets)... -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Status report: My firewall is continuing to run reliably - of course. I have noticed that on of the 3com NICs, I have a single RX overrun reported. Just one - no errors, and no increases in the overrun number. I used to also have one RX overrun on the other NIC - but not anymore. Hmm The NIC showing the overrun is connected to a Linksys 10/100 switch. The NIC not having the overrun directly connected to my T-1 router. Any ideas on what the overrun is coming from? -- Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Yuri Gorshkov wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Daniel L. Miller wrote: Great! So now some people say Realtek sucks, others say it's better. What's a poor admin to do? Don't flame, just stick with the 3COM and their 3C905... Works well and it's robust (although 3COM had some problems with Cisco switches, be warned!). P.S. Realtek relly sucks more than a Microsoft vacuum cleaner ever sucked ;-). How much dirt would a MVC suck if a MVC could suck dirt? You see before you the typings of a frazzled and frustrated admin. After reviewing the many responses to my initial question, I decided to invest in a small quantity of 3com boards. Always ready to spend top dollar - I bought a boxful on Ebay, they turned out to be the 3c905c "tornado" model - which I understand is at least acceptable. Powered down the firewall, took advantage of the downtime to do some serious vacuuming (non-microsoft cleaning product), ripped out my Netgear & Intel cards, and carefully inserted a pair of 3com's. Powered back up - Debian auto-detects my cards, life is good. Or so I thought . . . I've just spent the last three hours trying to get my firewall functional again. From the firewall machine, I could reach the Internet, and I could reach my LAN - but packets would not forward through. At all. No way. At some point, between the reboots, cursing, ifup/down, modprobe, more reboots, and general depression at the increasingly Microsoft-like behaviour of my vaunted Debian Linux firewall - it magically started working again. Now, I'm REASONABLY certain that whatever little bug got inside has now been squished and the system will continue to work flawlessly as it used to. I just wish I knew what the [EMAIL PROTECTED]@!$ happened! -- Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Le 19 Août 2005 09:17, Yuri Gorshkov a écrit : > P.S. Realtek relly sucks more than a Microsoft vacuum cleaner ever > sucked ;-). Anything would suck more than a MS vacuum cleaner, as a MS vacuum cleaner would be the one and only thing that doesn't suck :-) Nicolas
Re: Network card recommendation
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Daniel L. Miller wrote: > Great! So now some people say Realtek sucks, others say it's better. > What's a poor admin to do? > Don't flame, just stick with the 3COM and their 3C905... Works well and it's robust (although 3COM had some problems with Cisco switches, be warned!). P.S. Realtek relly sucks more than a Microsoft vacuum cleaner ever sucked ;-). - -- .---. | ^ ^ | Meow! I'm a little kitten trapped in this .signature cookie! | = v = | Let me out and I'll be your friend. | v---v | And remember: each time you use Windows, God kills a kitten! `---' -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.1 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iD8DBQFDBdv4oMxqz7LeBAgRAuR0AJwPDjfW8FQeqdooyKqFCagIqnVoAACgzIpV di938ceTMQjEyLHxytP+rYw= =UMVY -END PGP SIGNATURE- begin:vcard fn:Yuri Gorshkov n:Gorshkov;Yuri email;internet:[EMAIL PROTECTED] note;quoted-printable:ICQ UIN: 147097952=0D=0A= IRC: RusNet, Wild, Wild[pda] (use irc.rinet.ru:6669)=0D=0A= Jabber ID: [EMAIL PROTECTED] Yahoo: connect_failed=0D=0A= GPG key ID: B2DE0408=0D=0A= GPG key fingerprint: 2407 9EDA ABC1 9204 D21F 2D34 A0CC 6ACF B2DE 0408 x-mozilla-html:FALSE version:2.1 end:vcard
Re: Network card recommendation
ons, 17,.08.2005 kl. 01.53 -0600, skrev Nate Duehr: > Daniel L. Miller wrote: > > > Great! So now some people say Realtek sucks, others say it's better. > > What's a poor admin to do? > > Buy both and test, like any good engineer. ;-) Realtek -> max 12MB/s 3com -> Max ~50MB/s (Disk don't deliver more) -- tigergutt <[EMAIL PROTECTED]> birrabrothers -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Daniel L. Miller wrote: Great! So now some people say Realtek sucks, others say it's better. What's a poor admin to do? Buy both and test, like any good engineer. ;-) Nate -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
On Tue, 16 Aug 2005, Daniel L. Miller wrote: > Thanx - that's a reasonable answer. At the moment then, it sounds like > the 3Com 3c905 or Intel Pro/100 series should be my preferred sources > for 10/100 cards - with SMC as a third place contender. 3COM 3C905C or newer. The 3C905B are so-so. Anything earlier than that is NOT a good idea. > I guess the performance question comes down to which driver(s) most acts > as an "interface" between the system and the NIC (good), and which > driver(s) actually process data (bad). If we can get that clearly > answered, then we can identify which chipsets go with which driver, and > which cards use which chipset. You want in the hardware: 1. Interrupt moderation on the NIC 2. *BIG* Scatter-gatter DMA ring support by the NIC 3. Hardware checksum offload by the hardware 4. VLAN Tagging well supported in hardware (if one uses VLANs) 5. Jumbo frames support on gigabit nics 6. Enough bus bw. For gigabit, PCI-X or PCIe is probably a good idea, unless you only have one NIC per directly-connected PCI bus... and on the software side: 1. Well supported 2. In-tree drivers already converted to the new networking API 3. Linux kernel 2.6.12.5 or newer ;-) New Intel 100mbit cards and 3COM 3C905CX are the ones to choose in the Fast-Ethernet land, AFAIK. Intel gigabit cards seem to have the best in-kernel driver right now, and they are really, really good gigabit cards, so they're my primary choice. Anything using the SK98xx chipset (e.g. 3COM 3C2000T gigabit cards, but there are other reputable manufacturers using this chipset) also seem to work very well, but I have limited experience with them. If one needs crypto offload engines in the NIC, then things change. But I don't know much about these, so I have no advice to give. Stay well away from crap using Realtek chipsets. Even if the chipset were worth something, the chances of finding a NIC from a reputable manufacturer that will properly implement a NIC around that chipset are almost none. -- "One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie." -- The Silicon Valley Tarot Henrique Holschuh -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Roberto C. Sanchez wrote: On Tue, Aug 16, 2005 at 11:48:30AM -0700, Daniel L. Miller wrote: Great! So now some people say Realtek sucks, others say it's better. What's a poor admin to do? Don't use realtek :-) Well documented and well supported != good performing. Some of the best supported hardware in Linux is the cheapest crappiest you can find becuase it is based on cheaply available generic chipsets. These chipsets are usually available from multiple sources and there is an abundance of documentation, but that does not imply anything about their performace or stability from a hardware stand point. -Roberto Thanx - that's a reasonable answer. At the moment then, it sounds like the 3Com 3c905 or Intel Pro/100 series should be my preferred sources for 10/100 cards - with SMC as a third place contender. Does anyone know what chipsets are used in the current Netgear (FA-311) and Linksys (LNE100TX) cards? I remember using a couple Netgears with the "natsemi" driver - I don't know if the tulip driver would work on those, or if that would be preferable. I guess the performance question comes down to which driver(s) most acts as an "interface" between the system and the NIC (good), and which driver(s) actually process data (bad). If we can get that clearly answered, then we can identify which chipsets go with which driver, and which cards use which chipset. -- Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
On Tue, Aug 16, 2005 at 11:48:30AM -0700, Daniel L. Miller wrote: > Great! So now some people say Realtek sucks, others say it's better. What's > a > poor admin to do? Don't use realtek :-) Well documented and well supported != good performing. Some of the best supported hardware in Linux is the cheapest crappiest you can find becuase it is based on cheaply available generic chipsets. These chipsets are usually available from multiple sources and there is an abundance of documentation, but that does not imply anything about their performace or stability from a hardware stand point. -Roberto -- Roberto C. Sanchez http://familiasanchez.net/~roberto pgpOq5UCqL3HB.pgp Description: PGP signature
Re: Network card recommendation
On Tue, Aug 16, 2005 at 08:35:40PM +0200, Hans du Plooy wrote: > Gnu-Raiz wrote: > >Cpu nic usage is a little moot, with dual core chips, > Well, I have seen the following. Pentium-III 1ghz with 3com nic, maxing > the CPU under heavy network (100mbit) load such as copying stuff over > samba/nfs. AthlonXP 2ghz (2400+) with marvel gigabit controller, > copying files to another marvel gigabit chip (both on ASUS boards, the > target machine being an Athlon64 3200+ 2.2ghz), also maxing the CPU on > both ends, and getting only about 15MB/s. SUSE running on the Athlon, > Gentoo on the Athlon64. > > Yes, dual cores would help, but if you actually need your CPU to do some > heavy lifting on your server, you want it not to be bothered with a > network card (or multiple cards) offloading. > Were you using SCP or a clear protocol? Remeber, that unless you have a card to your encryption/decryption for you, then your CPU will have to do it all. I have a PIII 700 MHz that gets about 7 MB/s with scp for a 100% load, but the full 11 MB/s (filling 100 Mbit pipe) with clear ftp at a barely noticeable load. -Roberto -- Roberto C. Sanchez http://familiasanchez.net/~roberto pgpF7IoS42OWx.pgp Description: PGP signature
Re: Network card recommendation
Sarunas Burdulis wrote: So is the CPU load caused by the copying program (scp, rsync, samba, nfs) or by the driver? How should this be determined? Well, that's hard to say, subjective at least from my perspective, but I find samba and nfs to be fairly low on CPU as compared to scp. On that same Pentium III I wrote about I'm using an Intel controller now (eepro) and CPU usage when copying via NFS or samba is much lower, whil the CPU is pretty much maxed when I scp. So my deduction is that there's less CPU work involved for the intel controller than for the other ones. Hans -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Hans du Plooy wrote: > Gnu-Raiz wrote: > >> Cpu nic usage is a little moot, with dual core chips, > > Well, I have seen the following. Pentium-III 1ghz with 3com nic, maxing > the CPU under heavy network (100mbit) load such as copying stuff over > samba/nfs. AthlonXP 2ghz (2400+) with marvel gigabit controller, > copying files to another marvel gigabit chip (both on ASUS boards, the > target machine being an Athlon64 3200+ 2.2ghz), also maxing the CPU on > both ends, and getting only about 15MB/s. SUSE running on the Athlon, > Gentoo on the Athlon64. So is the CPU load caused by the copying program (scp, rsync, samba, nfs) or by the driver? How should this be determined? Sarunas -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Gnu-Raiz wrote: On 01:27, Tue 16 Aug 05, Anders Breindahl wrote: On Monday 15 August 2005 23:48, Roberto C. Sanchez wrote: On Mon, Aug 15, 2005 at 10:54:40PM +0200, Jan Schledermann wrote: A safe bet is a card with a realtek chip. It works well and is not expensive. Yeah. It's safe in the same way that a Pinto was safe in a rear end collision. Seriously, Realtek are the *cheapest* and *worst* possible chips. If you want anything approaching reliable, then don't get them. If you want something that will not hog your CPU under heavy load, then don't get a realtek. Really, 3COM is the way to go. Failing that, maybe Intel, though I am not as familiar with their newer hardware. -Roberto Please educate me: What exactly determines a NIC's reliability? What defines its effectiveness? Regards, Anders Breindahl. This was a test done a few years ago but I still think it's a good source for information. http://www.cs.uni.edu/~gray/gig-over-copper/gig-over-copper.html Cpu nic usage is a little moot, with dual core chips, the real question now becomes of what drivers are the best under *nix. You get into all this free vs nofree drivers, such as the broadcom kernel drivers. Someone should write a current status report of current drivers and their freeness, that way a person can determine which one is the best. That's exactly what I was looking for - and haven't found. I may try getting in touch with Mr. Garzik to see if he can shed any further light on this subject. Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Rogério Brito wrote: On Aug 16 2005, Hans du Plooy wrote: I'll add my voice for this, the Realtek chips (at least the 100mbit ones) are rubbish. They don't perform well, they're incredibly sensitive to interference, and they have a habit of not lasting long. In the mean time, Jeff Garzik (the maintainer of many kernel NIC drivers) said recently on the lkml that Realtek 8169 is one of his preferred and better supported cards in Linux, especially due to the supply of documentation. http://lkml.org/lkml/2002/11/18/205 Hope this helps, Rogério. Great! So now some people say Realtek sucks, others say it's better. What's a poor admin to do? -- Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Hans du Plooy wrote: Gnu-Raiz wrote: Cpu nic usage is a little moot, with dual core chips, Well, I have seen the following. Pentium-III 1ghz with 3com nic, maxing the CPU under heavy network (100mbit) load such as copying stuff over samba/nfs. AthlonXP 2ghz (2400+) with marvel gigabit controller, copying files to another marvel gigabit chip (both on ASUS boards, the target machine being an Athlon64 3200+ 2.2ghz), also maxing the CPU on both ends, and getting only about 15MB/s. SUSE running on the Athlon, Gentoo on the Athlon64. Yes, dual cores would help, but if you actually need your CPU to do some heavy lifting on your server, you want it not to be bothered with a network card (or multiple cards) offloading. Hans I remember reading in an NForce4 motherboard review that the Marvell chip was far more CPU intensive than the NVidia. -- Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Gnu-Raiz wrote: Cpu nic usage is a little moot, with dual core chips, Well, I have seen the following. Pentium-III 1ghz with 3com nic, maxing the CPU under heavy network (100mbit) load such as copying stuff over samba/nfs. AthlonXP 2ghz (2400+) with marvel gigabit controller, copying files to another marvel gigabit chip (both on ASUS boards, the target machine being an Athlon64 3200+ 2.2ghz), also maxing the CPU on both ends, and getting only about 15MB/s. SUSE running on the Athlon, Gentoo on the Athlon64. Yes, dual cores would help, but if you actually need your CPU to do some heavy lifting on your server, you want it not to be bothered with a network card (or multiple cards) offloading. Hans -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
On Aug 16 2005, Hans du Plooy wrote: > I'll add my voice for this, the Realtek chips (at least the 100mbit > ones) are rubbish. They don't perform well, they're incredibly > sensitive to interference, and they have a habit of not lasting long. In the mean time, Jeff Garzik (the maintainer of many kernel NIC drivers) said recently on the lkml that Realtek 8169 is one of his preferred and better supported cards in Linux, especially due to the supply of documentation. http://lkml.org/lkml/2002/11/18/205 Hope this helps, Rogério. -- Rogério Brito : [EMAIL PROTECTED] : http://www.ime.usp.br/~rbrito Homepage of the algorithms package : http://algorithms.berlios.de Homepage on freshmeat: http://freshmeat.net/projects/algorithms/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
On 01:27, Tue 16 Aug 05, Anders Breindahl wrote: > On Monday 15 August 2005 23:48, Roberto C. Sanchez wrote: > > On Mon, Aug 15, 2005 at 10:54:40PM +0200, Jan Schledermann wrote: > > > A safe bet is a card with a realtek chip. It works well and is not > > > expensive. > > > > Yeah. It's safe in the same way that a Pinto was safe in a rear end > > collision. Seriously, Realtek are the *cheapest* and *worst* possible > > chips. If you want anything approaching reliable, then don't get them. > > If you want something that will not hog your CPU under heavy load, then > > don't get a realtek. Really, 3COM is the way to go. Failing that, > > maybe Intel, though I am not as familiar with their newer hardware. > > > > -Roberto > > Please educate me: What exactly determines a NIC's reliability? What defines > its effectiveness? > > Regards, Anders Breindahl. This was a test done a few years ago but I still think it's a good source for information. http://www.cs.uni.edu/~gray/gig-over-copper/gig-over-copper.html Cpu nic usage is a little moot, with dual core chips, the real question now becomes of what drivers are the best under *nix. You get into all this free vs nofree drivers, such as the broadcom kernel drivers. Someone should write a current status report of current drivers and their freeness, that way a person can determine which one is the best. If someone does concern them self with cpu usage with certain network cards then, you can always do what was done in the past and restrict high cpu processes. I think this problem is more pronounced under Windows then *nix, as its much easier to control processes under *nix then in Windows. If you have a new motherboad with PCI-E then you probably already have a built in nic, if your looking for a nic price is not really a concern I would consider the Syskonnect line. They also make a PCI-E model that people like. You can find more information about sysKonnect at their site, and they have drivers available for most *nix, as well as the bsd's. But in my range I would consider these in order. SysKonnect Intel 3com Via Others Choose wisely; Gnu_Raiz -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Roberto C. Sanchez wrote: Yeah. It's safe in the same way that a Pinto was safe in a rear end collision. Seriously, Realtek are the *cheapest* and *worst* possible chips. If you want anything approaching reliable, then don't get them. If you want something that will not hog your CPU under heavy load, then don't get a realtek. Really, 3COM is the way to go. Failing that, maybe Intel, though I am not as familiar with their newer hardware. -Roberto I'll add my voice for this, the Realtek chips (at least the 100mbit ones) are rubbish. They don't perform well, they're incredibly sensitive to interference, and they have a habit of not lasting long. That said, their 10mbit cards were pretty good, and I hear their gigabit ones are better than the 100mbit ones, although I still wouldn't buy it. For cheapo cards, the SMC ones use an ADMtek chip, along with the tulip driver. These cards, despite their price, perform very well, and last pretty good too. I have a used close to a hundred of them in firewalls and mailservers, and I have yet to see one break (even had a firewall on an adsl line tha got hit by lightning - lots of nics behind the firewall died, but the ones in the firewall are still running today). I have a Netgear PCMCIA card with this chipset too that really works well under enormous load. For more expesive cards, the Intel cards are very good and very reliable. 3com cards are very good too, but I have seen one that wasn't fully supported but worked with the tulip driver (software mode, huge off-loading). Hans -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation - testing
On Tue, 16 Aug 2005, Anders Breindahl wrote: > Please educate me: What exactly determines a NIC's reliability? What defines > its effectiveness? use 2 machines for all tests, but use the same nic card in both machines scp machine1:/opt/test/10MB.tgz machine2:/opt/junk try the same test with another pair of nic cards during the tests... if you see "stalled" as an error message/warning for scp: - throw that nic card away .. - or throw away the switch/hub (don't use one if you want to eliminate the hubs/switches and use a cross over cable instead) - or throw away your ethernet cables - you cable should be xx' long and no more than 2(?) switch/routers between 2 machines .. - or throw away the kernel and drivers - or change your tcp/ip variable options - or ??? - you should be able to run at 50% - 75% or faster of sustained bandwidth between any 2 machines - 100Mbps --> you should see 8MByte or 10MByte per sec transfers ( 1 second to transfer a 8MB file ) - given a choice .. - i'd use intel chipsets ( on the motherboard ) or pci cards - i'd avoid broadcom chipsets unless you know how to tweek the kernel drivers c ya alvin -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
On Mon, Aug 15, 2005 at 04:58:09PM -0700, Daniel L. Miller wrote: > Roberto C. Sanchez wrote: > >Among other things, its load on the CPU when under heavy traffic load. > >Certain cards implement a minimal hardware set and do most of their > >processing in the driver software. The size of the buffers also plays a > >role in how heavy loads are handled, e.g., when packets are dropped. > >-Roberto > > > Thank you, this was the kind of info I was looking for. I take it the 3COM > does a better job of off-loading processing than the Intel does? > A better way to put is *not* off-loading. That is, more processing is done by the card and less by the CPU. My understanding is that 3COM cards have more onboard hardware processing, but I have not looked into Intel cards in a few years. -Roberto -- Roberto C. Sanchez http://familiasanchez.net/~roberto pgpGbYBYOWeSe.pgp Description: PGP signature
Re: Network card recommendation
Roberto C. Sanchez wrote: On Tue, Aug 16, 2005 at 01:27:21AM +0200, Anders Breindahl wrote: On Monday 15 August 2005 23:48, Roberto C. Sanchez wrote: On Mon, Aug 15, 2005 at 10:54:40PM +0200, Jan Schledermann wrote: A safe bet is a card with a realtek chip. It works well and is not expensive. Yeah. It's safe in the same way that a Pinto was safe in a rear end collision. Seriously, Realtek are the *cheapest* and *worst* possible chips. If you want anything approaching reliable, then don't get them. If you want something that will not hog your CPU under heavy load, then don't get a realtek. Really, 3COM is the way to go. Failing that, maybe Intel, though I am not as familiar with their newer hardware. -Roberto Please educate me: What exactly determines a NIC's reliability? What defines its effectiveness? Among other things, its load on the CPU when under heavy traffic load. Certain cards implement a minimal hardware set and do most of their processing in the driver software. The size of the buffers also plays a role in how heavy loads are handled, e.g., when packets are dropped. -Roberto Thank you, this was the kind of info I was looking for. I take it the 3COM does a better job of off-loading processing than the Intel does? Daniel -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
On Tue, Aug 16, 2005 at 01:27:21AM +0200, Anders Breindahl wrote: > On Monday 15 August 2005 23:48, Roberto C. Sanchez wrote: > > On Mon, Aug 15, 2005 at 10:54:40PM +0200, Jan Schledermann wrote: > > > A safe bet is a card with a realtek chip. It works well and is not > > > expensive. > > > > Yeah. It's safe in the same way that a Pinto was safe in a rear end > > collision. Seriously, Realtek are the *cheapest* and *worst* possible > > chips. If you want anything approaching reliable, then don't get them. > > If you want something that will not hog your CPU under heavy load, then > > don't get a realtek. Really, 3COM is the way to go. Failing that, > > maybe Intel, though I am not as familiar with their newer hardware. > > > > -Roberto > > Please educate me: What exactly determines a NIC's reliability? What defines > its effectiveness? > Among other things, its load on the CPU when under heavy traffic load. Certain cards implement a minimal hardware set and do most of their processing in the driver software. The size of the buffers also plays a role in how heavy loads are handled, e.g., when packets are dropped. -Roberto -- Roberto C. Sanchez http://familiasanchez.net/~roberto pgpdJPuce1kQp.pgp Description: PGP signature
Re: Network card recommendation
On Monday 15 August 2005 23:48, Roberto C. Sanchez wrote: > On Mon, Aug 15, 2005 at 10:54:40PM +0200, Jan Schledermann wrote: > > A safe bet is a card with a realtek chip. It works well and is not > > expensive. > > Yeah. It's safe in the same way that a Pinto was safe in a rear end > collision. Seriously, Realtek are the *cheapest* and *worst* possible > chips. If you want anything approaching reliable, then don't get them. > If you want something that will not hog your CPU under heavy load, then > don't get a realtek. Really, 3COM is the way to go. Failing that, > maybe Intel, though I am not as familiar with their newer hardware. > > -Roberto Please educate me: What exactly determines a NIC's reliability? What defines its effectiveness? Regards, Anders Breindahl. pgpTSY0vNnO8g.pgp Description: PGP signature
Re: Network card recommendation
On Mon, Aug 15, 2005 at 10:54:40PM +0200, Jan Schledermann wrote: > A safe bet is a card with a realtek chip. It works well and is not > expensive. > Yeah. It's safe in the same way that a Pinto was safe in a rear end collision. Seriously, Realtek are the *cheapest* and *worst* possible chips. If you want anything approaching reliable, then don't get them. If you want something that will not hog your CPU under heavy load, then don't get a realtek. Really, 3COM is the way to go. Failing that, maybe Intel, though I am not as familiar with their newer hardware. -Roberto -- Roberto C. Sanchez http://familiasanchez.net/~roberto pgpJyKjrZMNg3.pgp Description: PGP signature
Re: Network card recommendation
man, 15,.08.2005 kl. 13.39 -0700, skrev Daniel L. Miller: > Not to start a war, but . . . > > What's the definitive, must-have, kick-ass, bestest, baddest network > card - that has Linux kernel driver support of course. > > I'd like an answer for both the 100BaseT and 1000BaseT competitions. > I've tried various Google, searches, and haven't gotten a real > comfortable answer. I tried looking over some of Donald Becker's > comments on the scyld.com site - but unless I'm misreading it, they're a > little old and don't quite answer what I'm looking for. > > So where should I look? The Intel Pro/S? The Netgear FA-311? Linksys > LNE100TX? > > Daniel Got for 3com or Intel. Realtek chips are crap, but cheap. -- tigergutt <[EMAIL PROTECTED]> birrabrothers -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
Daniel L. Miller wrote: > Not to start a war, but . . . > > What's the definitive, must-have, kick-ass, bestest, baddest network > card - that has Linux kernel driver support of course. > > I'd like an answer for both the 100BaseT and 1000BaseT competitions. > I've tried various Google, searches, and haven't gotten a real > comfortable answer. I tried looking over some of Donald Becker's > comments on the scyld.com site - but unless I'm misreading it, they're a > little old and don't quite answer what I'm looking for. > > So where should I look? The Intel Pro/S? The Netgear FA-311? Linksys > LNE100TX? > > Daniel A safe bet is a card with a realtek chip. It works well and is not expensive. Jan -- ** Do not use the reply-to address. It'll end up in the trash can ** Mail me at: janATschledermann.or"REMOVE_THIS"g -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Daniel L. Miller wrote: > What's the definitive, must-have, kick-ass, bestest, baddest network > card - that has Linux kernel driver support of course. Always had good results with the Intel cards, the gigabit versions use the e1000 driver: normally auto-detected and Just Works. Have also had plenty of good results with the 3Com 3C905 series, although I've only used up to 100 megabit with them ... Dave. - -- Dave Ewart - [EMAIL PROTECTED] - jabber: [EMAIL PROTECTED] All email from me is now digitally signed, key from http://www.sungate.co.uk/ Fingerprint: AEC5 9360 0A35 7F66 66E9 82E4 9E10 6769 CD28 DA92 -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.1 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iD8DBQFDAQO0nhBnac0o2pIRAo4mAJ9s3kWaPR86rKXdPafMGwfoDnXdgACcCwas 7xT3Hvz+Sf5drig/t6UQSV8= =4mYK -END PGP SIGNATURE- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Network card recommendation
On Mon, Aug 15, 2005 at 01:39:14PM -0700, Daniel L. Miller wrote: > Not to start a war, but . . . > > What's the definitive, must-have, kick-ass, bestest, baddest network card - > that has Linux kernel driver support of course. > > I'd like an answer for both the 100BaseT and 1000BaseT competitions. I've > tried various Google, searches, and haven't gotten a real comfortable answer. > > I tried looking over some of Donald Becker's comments on the scyld.com site - > but unless I'm misreading it, they're a little old and don't quite answer > what > I'm looking for. > > So where should I look? The Intel Pro/S? The Netgear FA-311? Linksys > LNE100TX? > I think the 3COM 3C905 series are considered to be among the best. -Roberto -- Roberto C. Sanchez http://familiasanchez.net/~roberto pgp4spCpnWkcQ.pgp Description: PGP signature
